Abstract: With the rapid development of city railroad vehicles and bullet train, the automaticwelding technology should be improved instantly.Welds are numerous and short in sliding door and welding should be high quality and exact, so a vision tracking system of discontinuous robot arc welding with normal lamp-house was produced. A series of processes of vision tracking image with normal lamp-house were established, which is the image processing technologies that gray-scale transform, median filter, image segmentation, etc.Traditional image segmentation includes gray-scale transform and line thinning, etc.They seam position can be found by its sides.However, the result by this method is easily disturbed by noises such as dark or bright cutter traces, and so on.A movable window technology was produced.This technology takes the positio n, which the average gray scale of all the pixels in a window is the minimum one, as the current position.And the window has the same size and shape as the seam and can be moved through the acquired image.The technology takes the whole pixels of the seam as the calculating data.As a result, it is powerful anti-jamming and can effectively filter the noises of the dark and bright cutter traces.The error of the seam position can be directly calculated after the current and standard positions are compared.The actualizing steps of movable window technology are also presented.The tests at arc welding robot workstation for the sliding door show that the seam image processing system has powerful adaptability and anti-jamming ability.And its fast work speed can meet the request of real time tracking.